Optimization Apparatus Using Content-Addressable Memory for Weighting Coefficients

Solution Overview

Problem

As the number of state variables increases in optimization problems, the number of weighting coefficients grows rapidly, leading to increased storage capacity requirements and reduced calculation speeds in existing optimization apparatuses, particularly when implemented with semiconductor integrated circuits.

Innovation Solution

The optimization apparatus employs a configuration where weighting coefficients between non-interacting neurons are not stored, using a content-addressable memory (CAM) to reduce storage capacity and enhance calculation speed by only storing coefficients for interacting neurons, allowing for larger-scale optimization problems to be solved on a single chip.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If the number of state variables increases in optimization problems, then the problem scale and complexity increase, but the number of weighting coefficients grows rapidly, leading to increased storage capacity requirements and reduced calculation speeds

Engineering Contradiction:

Improvenumber of state variablesVSAvoidstorage capacity

Core Design Contradiction:

Quantity of substanceVSVolume of stationary object

Solution Approach 1:

The patent extracts only the necessary weighting coefficients for interacting neurons into the storage unit, excluding coefficients for non-interacting neurons. This selective extraction reduces the storage capacity requirement from O(n²) to O(m) where m is the number of interacting neuron pairs, directly resolving the contradiction between problem scale and storage requirements.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent segments the weighting coefficient matrix into interacting and non-interacting portions, storing only the relevant interacting coefficients in a dedicated storage unit. This segmentation approach allows the system to handle larger optimization problems by reducing the storage burden from storing all possible neuron pairs to only those with actual interactions.

Inventive Principle:

Principle #1Segmentation

2Quantity of substance

If the number of weighting coefficients increases, then the problem scale increases, but memory access delays increase, reducing calculation speed

Engineering Contradiction:

Improvenumber of weighting coefficientsVSAvoidcalculation speed

Core Design Contradiction:

Quantity of substanceVSSpeed

Solution Approach 1:

The patent extracts only the necessary weighting coefficients for interacting neurons into the storage unit, excluding coefficients for non-interacting neurons. This selective extraction reduces the storage capacity requirement from O(n²) to O(m) where m is the number of interacting neuron pairs, directly resolving the contradiction between problem scale and storage requirements.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent pre-identifies and stores only the weighting coefficients for interacting neurons before the optimization calculation begins. This preliminary action of selective storage eliminates the need to access and process zero or insignificant coefficients during calculation, thereby reducing memory access delays and improving calculation speed for large-scale problems.

Inventive Principle:

Principle #10Preliminary action

3Loss of information

If all weighting coefficients are stored in memory, then complete neuron interaction data is available, but storage capacity requirements and memory access delays increase

Engineering Contradiction:

Improveneuron interaction data completenessVSAvoidstorage capacity

Core Design Contradiction:

Loss of informationVSVolume of stationary object

Solution Approach 1:

The patent extracts only the necessary weighting coefficients for interacting neurons into the storage unit, excluding coefficients for non-interacting neurons. This selective extraction reduces the storage capacity requirement from O(n²) to O(m) where m is the number of interacting neuron pairs, directly resolving the contradiction between problem scale and storage requirements.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent applies local quality by storing weighting coefficients selectively based on their significance - only interacting neurons have their coefficients stored in the storage unit, while non-interacting neurons are excluded. This localized storage approach maintains information completeness for relevant interactions while minimizing storage capacity requirements.

Inventive Principle:

Principle #3Local quality

Data Source

PatentUS11526740B2Optimization apparatus and optimization method

Publication Date: 2022.12.13 FUJITSU LTD

AI summary

An optimization method includes holding combining destination information indicating a combining destination neuron to be combined with a target neuron which is one of a plurality of neurons corresponding to a plurality of spins of an Ising model obtained by converting an optimization problem, the target neuron being different in a plurality of neuron circuits; holding a weighting coefficient indicating a strength of combining between the target neuron and the combining destination neuron, and outputting the weighting coefficient corresponding to the combining destination information; permitting an update of a value of the target neuron by using the weighting coefficient output and the value of the update target neuron, and outputting a determination result indicating whether or not the value of the target neuron is permitted to be updated; and determining the update target neuron based on the plurality of determination results respectively output and outputting the update target information.